Large area flat panel imagers function by accumulating charge on capacitors generated by pixels of p-i-n photodiodes (amorphous silicon or organic semiconductor) with scintillators or by pixels of photoconductors. Typically, many pixels are arranged over a surface of the imager where TFTs (or single and/or double diodes) at each pixel connect the charged capacitor to a read out amplifier at the appropriate time. A pixel is composed of the scintillator/photodiode/capacitor/TFT or switching-diode combination or by the photoconductor/capacitor/TFT or switching-diode combination. Often the photodiode intrinsically has enough capacitance that no separate charge storage capacitor is required. As illustrated in FIG. 1A, radiation (e.g., alpha, beta, gamma, X-ray, neutrons, protons, heavy ions, etc.) strikes the scintillator and causes the scintillator to generate visible light. The visible light strikes a photodiode and generates an electric current. Alternatively, an imager may be configured such that the radiation strikes a biased photoconductor to generate the electric current, as illustrated in FIG. 1B. The current charges a capacitor (where the illustrated capacitor includes the self capacitance of the photoconductor) and leaves a charge on the capacitor. The integrated charge on the capacitor is proportional to the integrated light intensity striking the respective photoconductor for a given integration time. At an appropriate time, a switch (e.g., a TFT or switching diode(s)) activates and reads out the charge from the capacitor onto a charge sensitive amplifier (not shown).
For long integration times, typically over 20 seconds for amorphous silicon technology, there is a linear increase in charge QLi to the capacitor charge (in coulombs) of pixel “i,” as a function of discrete frame time T, due to a constant leakage (or dark) current from the switch (e.g., TFT), diode or photodetector. This dark current ID is on the order of 1-2 femtoamps (fA) for 100 to 200 micron wide pixels of amorphous silicon TFT construction. The expression for QLi is the dark current ID multiplied by T. Dark current ID may be constant or time varying; giving an excess charge QE contribution that is either linear or non-linear with respect to time, respectively. This linear dark charge contribution to QLi is subtracted from the total charge QTi read off the capacitor of pixel “i” in order to provide the true image charge QSi. Subtracting the dark-current charge contribution (either linear or non linear), from the total charge QTi read off the capacitor, is called background (or offset) correction.
In addition to dark current charge contributions from the switch (e.g., TFT) there are leakage (or dark) current charge contributions from the capacitor and the photodiode. The true image charge QSi is obtained by subtracting the background (or offset) dark charge contributions from the measured charge QTi of pixel “i.” The simplest background correction method is to subtract a constant fraction of the charge that was present on pixel “i” during at least one, and sometimes additional, prior frames.
Prior background correction methods have been implemented to estimate offset correction. One prior background correction method discussed in U.S. Pat. No. 5,452,338, isolates the offset image by acquiring an image when the detector is not exposed to X-rays, using the same timing used in acquiring the X-ray exposed images. The image acquired after exposure is then subtracted from future frames. One problem with the use of one single frame in determining the offset correction is the offset image introduces additional noise. To reduce the additional noise, multiple non-exposed images may be acquired and then averaged. One problem using a single image or an average of multiple images is the offset signals may drift with time, temperature, and other extrinsic factors while the single image or averaged image remains constant.
Another prior background correction method discussed in a paper by Sussan Pourjavid, et al., entitled “Compensation for Image Retention in an Amorphous Silicon Detector” (SPIE Conference on Physics of Medial Imaging, February 1999), and U.S. Pat. No. 5,452,338, continuously updates the offset images to compensate for drift in the offset signals. The method, described in the above references, models the time response of the background contribution from leakage current (or dark image) in the diodes as a linear time variant system (LTI) using linear systems theory (least square method). The LTI system derived from the response model is then used to predict the offset needed for image correction. However, in modern medical imaging equipment, for example, there is a demand for real-time, 30 frames per second images, where scans are made with 33 millisecond integration times. Even more advanced imaging applications, like computed tomography, can use even higher frames rates of 120, 360 or even 900 frames per second, corresponding to respective integration times of 8, 3 and 1 milliseconds, respectively. Background correction using least square prediction of discrete frame time in such situations is not as effective. With near real-time imaging, for example 3 frames per second (FPS) or faster, background correction with least square prediction can introduce significant image errors and artifacts. A more effective method for background (or offset) correction is needed in short-integration-time imaging applications.